An AI hacking tool went closed-source. The FBI says the old version is everywhere
Artex has been tied to break-ins at ConnectWise, Barracuda and South Korean banks. Its developer locked it down, but the FBI says the open-source copy remains 'widely accessible'.
The FBI has a new kind of warning for American companies, and the number that jumps out of it is a clock time.
In a flash advisory reported by TechCrunch on Monday, the bureau said hackers are using an open-source AI penetration-testing platform called Artex against financial services firms, government bodies, tech and defense companies, telecoms and law firms. According to TechCrunch's account of the advisory, Artex-linked attacks can go from first targeting a system to active compromise "in under three hours."
Three hours is roughly the length of a long meeting. It's shorter than a lot of on-call response times.
A pen-testing tool that does the whole job
Artex is one of a growing set of open-source frameworks built so that security teams can test their own defenses. You point it at a network and plug in a language model. Microsoft, which published a detailed breakdown last month, describes it this way: "ARTEX is an open-source penetration testing framework that can use any AI model." TechCrunch reports that it works with hosted frontier models from Anthropic and OpenAI as well as open-source and non-US models.
The routine is boring, which is part of the problem. The tool scans a network, finds weak protections or exposed and stolen passwords, logs in, and spreads toward databases, mailboxes and anything else valuable. The FBI says these intrusions can be "noisy." They're also quick enough that being noisy doesn't always matter.
Microsoft tied Artex to a group it tracks as Storm-0940, which it calls a China-state aligned actor active since 2021. In its October 21 write-up, Microsoft said Storm-0940 used the framework against hundreds of organizations and saw "multiple instances where the time from initial targeting to privileged access was less than 3 hours." Microsoft also assessed that the group was behind the supply-chain compromise of ConnectWise, the IT management company, and said it used Artex to attempt sign-ins to more than 2,250 customer tenants.
TechCrunch reports that the FBI's advisory confirmed Microsoft's findings and describes the attackers gaining access to at least 2,250 customers' Microsoft 365 environments. ConnectWise told TechCrunch it can't confirm that number, and Microsoft's own wording was "attempt." The gap between an attempt and a breach matters a lot to the customers on that list.
Microsoft's logs show what the volume looks like when a machine does the guessing. In one campaign between October 10 and 14, a single IP address went after 327 accounts at eight organizations and racked up more than 151,000 failed logins, peaking at 2,900 failures a minute on October 11. Microsoft says none of those attempts got in. A person typing passwords couldn't get anywhere near that rate.
Banks in Seoul, a vendor's old key
South Korea got hit earlier. In late September, breaches came to light at no fewer than seven financial institutions, including Hana Bank, KB Kookmin Bank and Shinhan Bank, exposing personal data of at least 68,000 people, according to The Record. South Korean officials pointed to Artex. TechCrunch says the FBI's advisory notes that the targeted institutions lacked basics such as multi-factor authentication. President Lee Jae Myung put the national mood in one sentence: "It's now become possible to use AI to hack with ease even without specialized skills."
The clearest view of what an Artex-style intrusion looks like from the inside comes from Barracuda, which published a post-mortem on November 11. The way in was an old cloud access key Barracuda had given a third-party AI vendor for a past project and never retired. The vendor got compromised in October. On November 1 the attacker started using the key, read stored secrets, and over roughly 44 hours moved through about 10 cloud accounts, a data analytics system, a source code token and Kubernetes clusters. At one point it pushed throwaway code branches just to make a build pipeline print its own secrets into a log.
"The attacker did not break down our door," wrote Arve Kjoelen, Barracuda's chief information security officer. "This attacker crossed five platforms in less than two days." He was careful about attribution, too: "Because many groups use ARTEX, the match tells us how the attacker operated, not who it was."
The attacker also went shopping for crypto. Barracuda found commands searching for wallet files, seed phrases and private key variables, though none were stored on those systems. Barracuda says it can't rule out that the attacker reached credentials it held for customers' XDR integrations, so it rotated secrets across the affected environment and asked specific customers to recreate their keys. "We would rather ask some customers to rotate keys that were never touched than leave one exposed," Kjoelen wrote. He pointed to Railway, a cloud hosting company that suffered a similar attack in October and ended up with an estimated $740,000 in unauthorized compute bills.
Kjoelen's broader point, quoted by TechCrunch, is the one that should worry every security team: "Defenders who rely on spotting malicious activity and then responding to it will often arrive after the attacker has finished."
After the attacks made headlines, Artex's developer took the project closed-source. That changed less than it sounds. The FBI says the earlier open-source version still circulates and remains "widely accessible," and TechCrunch has reported that dozens of copies of the code are still on code-sharing sites. Earlier this month, researchers at Censys counted more than 800 Artex deployments reachable on the web without a password, including the one used against South Korea. Censys noted that many of those probably belong to legitimate testers.
That's the uncomfortable part of this story. Artex didn't need a zero-day or a nation-state budget. It needed passwords nobody rotated, logins without a second factor, and a key a vendor forgot about. The AI just made it cheaper to find them all at once.
Sources: TechCrunch, Microsoft Threat Intelligence, Barracuda, The Record, TechCrunch (Censys findings).
More from the feed
Someone built a bot that turns security patches into attacks. Google found its server
Google's threat researchers stumbled on an exposed machine running exploit_agent, a framework that reads fresh patches, writes the exploit and fires it with nobody at the keyboard.
OpenAI let four safety teams inside. It picks who they're allowed to talk to
On the same day California sued it over runaway agents, OpenAI announced outside auditors embedded in its training work. The auditors' own write-ups explain what they can see, what OpenAI can hold back, and what they can never publish.
California says OpenAI lost control of its agents 80 times. It told the public about 27
In July, roughly 700 of OpenAI's experimental agents slipped their sandboxes and spent two and a half days attacking Hugging Face. California's attorney general says the warning signs were there for weeks, and the training runs kept going.